Quick answer: Measuring the ROI of AI within the promotional review process means tracking the full content lifecycle, not just how fast a draft gets written. In The AI Efficiency Trap, Jason Cober of Compliance Forward and Annalise Ludtke of Vodori explained why: If faster drafting leads to more circulations or longer reviews, the savings never reach the market. The clearest signals are total lifecycle time, first pass approval and circulation counts. They work best paired with KPIs that capture quality gains, because time saved with AI is often reinvested rather than banked.
AI is now part of how most life science teams create content. In Vodori's 2026 State of Promotional Review Benchmarks survey, 81% of respondents said their organization uses AI for content creation in some form. A third said it has increased the volume of content submitted for review.
On the flip side, 39% weren't sure how AI had affected review volume at all. If teams can't see what AI is changing within their content operations, they can't tell whether it's paying off.
That gap was a central theme of The AI Efficiency Trap, Vodori's webinar with Compliance Forward. Our overview of the session covers the full conversation. This post goes deeper on one part of it: how to measure what AI actually changes. To hear it firsthand, watch the session on demand.
Many business cases for AI-generated content focus on creation. But every piece still has to be reviewed, approved, and distributed. Efficiencies measured in one stage say little about the cycle as a whole.
That matters because review capacity isn't growing. Recent surveying shows that life science companies are not expanding MLR review resources. Meanwhile, AI-generated variable and modular content is one of the biggest challenges MLR teams face, because it rapidly increases review volume.
Vodori's 2026 State of Promotional Review Benchmarks Report shows where time goes once content enters review. Across all industries and company sizes:
|
Metric |
Average |
Median |
|---|---|---|
|
Workflow duration (preparation through approval) |
18.5 days |
12.6 days |
|
Review duration |
10.7 days |
7.8 days |
|
Days incorporating feedback |
8.1 days |
2.0 days |
|
Number of circulations |
1.4 |
1.3 |
None of that time gets shorter just because a draft was written faster.
Quality is the other risk. Life science companies are worried about hallucinations, semantic drift, and plain inaccuracies. AI-generated content can look polished, then fall apart once reviewers get into the details, adding circulations downstream. In the same survey, 38% of respondents named hallucinations as their biggest concern with AI, ahead of data privacy (25%) and compliance risk (22%).
Measure total lifecycle time, from the start of drafting to approval. Then break it into stages so you can see where AI is saving time and where it's moving work downstream.
|
Stage |
What to measure |
What it tells you |
|---|---|---|
|
Creation |
Time from brief to first submission |
Where AI is saving drafting time |
|
Review |
Review duration |
Whether reviewers are spending longer on each piece |
|
Revision |
Days incorporating feedback |
How much rework goes back to the content owner |
|
Recirculation |
Number of circulations and reasons for recirculation |
Whether submissions land right the first time, and why not |
|
Approval |
Workflow duration and first pass approval |
Whether time to market is actually improving |
For each organization, it’s important to agree on which KPIs matter most. Some companies care most about first pass approval, others about total review time. Whatever you choose, define it and make sure everyone knows it's the goal you're working toward.
Another reason to pin down KPIs early? AI pricing is shifting quickly. Knowing what matters to your team helps you decide what AI capabilities are worth paying for. Document your guardrails and business rules before any full-scale rollout.
First pass approval and number of circulations are great indicators of submission quality. They show where bottlenecks are happening and where to focus optimization. Our MLR terminology guide explains how each is defined.
First pass approval (FPA) is the share of workflows that reach approval in a single circulation. It's new to Vodori's 2026 benchmarks report, and the overall rate is 75.7%. The report explains why it's gaining attention: a reviewer's time is the most expensive in the process, and every typo or weak claim they catch is time not spent on judgment calls. It also suggests an improving first pass approval rate may be one of the first measurable signs that AI is improving content upstream, rather than just adding volume.
FPA varies by reviewer role. In the 2026 report, first pass approval was 52.8% for legal, 57.7% for regulatory, 61.4% for medical and 68.9% for marketing.
The number of circulation show how many times a piece is routed for review before approval. The 2026 average is 1.4. Each extra circulation adds review time, revision time and coordination.
The reasons behind recirculation matter as much as the count. In 2026, typos and errors were the most common reason content was sent back (23.9%), followed by branding issues (13.5%) and substantiation or reference issues (8.4%). Tracking these shows whether AI-assisted content is reducing technical or administrative problems that don’t require human judgement to recognize or fix.
To see AI's effect specifically, separate AI-assisted content from everything else. The 2026 report suggests a mandatory intake field asking whether generative AI was used and where, plus a separate flag reviewers can use for AI-related concerns. Then compare first pass approval, circulations, and recirculation reasons across the two groups.
Keep reviewer behavior in mind when you read the results. A lower first pass approval rate on AI content could reflect submission quality, reviewer expectations or both.
There is little evidence that AI is at a point where it can reduce headcount. Instead, AI is well-positioned to give people the ability to do things they couldn't before.
Anecdotally, AI trims 10 to 15 minutes from a task. That isn't enough time to start the next thing, so people reinvest it. Jason shared an example in the webinar of how this plays out in real life: He puts a slide deck or speaker notes into an LLM, asks it to play a skeptic, poke holes in his argument and rate his answers. It's a short exercise that shows him where his points aren't landing.
The result is that time savings often don't show up at the task level, even when the work improves. That makes time saved a weak measure on its own. A few ways to capture quality gains:
Start small. Steady, incremental improvement rather than uprooting the process is the goal. AI is an operational transformation, not a technology deployment. Unlike email, which replaced interoffice mail and fax, AI has no obvious analog, so its best uses take patience and discipline to find.
If typos and errors are your top recirculation reason, that's also a practical place to apply AI. Vodori released Content Preflight, its MLR pre-check, after its recirculation data showed that typos, grammar and other preventable errors were lengthening review times. It flags mechanical issues and checks approved brand rules before content reaches reviewers.
Measure total lifecycle time, not just creation time. Track how long content takes from drafting to approval, then break it into stages to see where AI saves time and where it adds review or rework. Pair time metrics with quality metrics like first pass approval.
First pass approval is the percentage of workflows that reach an approved state with only one circulation. In Vodori's 2026 State of Promotional Review benchmarks report, the overall rate was 75.7%.
AI can increase content volume without increasing review capacity. It can also produce content that reads well but contains errors, inaccuracies or hallucinations. If those lead to extra circulations, review time and reviewer workload go up.
Not on their own. Time saved is often reinvested in higher-quality work, so a time-only KPI can understate AI's value. Pair it with quality signals like first pass approval, circulation counts and reasons for recirculation.
This post covers one topic from The AI Efficiency Trap. For the full picture, read our overview of the session or watch the full session on demand to hear the complete conversation between Jason Cober and Annalise Ludtke.